JP7634356B2 - Medical information processing equipment - Google Patents

Description

The embodiments disclosed in this specification and the drawings relate to a medical information processing device.

Conventionally, there is known a technology called Clinical Decision Support Systems (CDSS) that supports doctors in making diagnoses when treating patients. Generally, CDSS uses a trained model that presents information about a patient's disease and treatment methods for that disease based on the results of learning data based on past performance through machine learning or the like. In addition, even after the start of operation, CDSS may update the model by learning based on new training data.

International Publication No. 2017/081984

One of the problems that the embodiments disclosed in this specification and the drawings attempt to solve is to support, in the learning of a clinical decision support system, the preferential improvement of diagnostic accuracy for cases that are important when a doctor makes a diagnosis. However, the problems that the embodiments disclosed in this specification and the drawings attempt to solve are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems.

A medical information processing apparatus according to an embodiment includes an acquisition unit, a case extraction unit, and an output control unit. The acquisition unit acquires a first diagnosis result for each of a plurality of diagnosed patients by the clinical decision-making support system before learning the first data, a second diagnosis result for each of the plurality of diagnosed patients by the clinical decision-making support system after learning the first data, a doctor's diagnosis result for each of the plurality of diagnosed patients, and first data. The case extraction unit calculates an update influence index and a diagnosis influence index for each of a plurality of cases included in the first data based on the first diagnosis result, the second diagnosis result, and the doctor's diagnosis result, and extracts, as presentation target cases, cases that are affected by the update of the clinical decision-making support system due to learning the first data, whose update influence index is equal to or greater than a first specified standard and whose diagnosis influence index is equal to or less than a second specified standard, from among the plurality of cases included in the first data. The output control unit causes the output unit to output the presentation target cases in association with the diagnosis influence index and the update influence index , which indicate the degree of influence that the clinical decision-making support system has on the doctor's diagnosis in the presentation target cases. The first data includes a plurality of cases in which information on a disease for each of a plurality of diagnosed patients is associated with a true value representing a treatment result for each of the plurality of diagnosed patients. The update influence index is an index based on a difference between the first diagnosis result and the true value and a difference between the second diagnosis result and the true value. The diagnosis influence index is an index based on a difference between the first diagnosis result and a doctor's diagnosis result.

FIG. 1 is a diagram illustrating an example of the overall configuration of an information processing system according to the first embodiment. FIG. 2 is a diagram showing an example of a decision tree used in the CDSS according to the first embodiment. FIG. 3 is a diagram illustrating an example of a change in output before and after updating the CDSS according to the first embodiment. FIG. 4 is a diagram showing an example of information acquired by the medical image processing apparatus according to the first embodiment. FIG. 5 is a diagram showing an example of a relationship between a diagnosis result of the CDSS and a diagnosis result of a doctor according to the first embodiment. FIG. 6 is a diagram showing another example of the relationship between the diagnosis result of the CDSS and the diagnosis result of a doctor according to the first embodiment. FIG. 7 is a diagram showing yet another example of the relationship between the diagnosis result of the CDSS and the diagnosis result of a doctor according to the first embodiment. FIG. 8 is a diagram illustrating an example of a method for extracting an influential variable according to the first embodiment. FIG. 9 is a diagram illustrating an example of the analysis screen according to the first embodiment. FIG. 10 is a diagram illustrating an example of adding a new case according to the first embodiment. FIG. 11 is a diagram illustrating an example of classification of variables according to the first embodiment. FIG. 12 is a diagram illustrating an example of the learning process according to the first embodiment. FIG. 13 is a flowchart showing an example of the flow of the update support process according to the first embodiment. FIG. 14 is a diagram illustrating an example of an analysis screen according to the second embodiment. FIG. 15 is a diagram showing an example of an analysis screen according to the first modification.

Below, an embodiment of the medical information processing device will be described in detail with reference to the drawings.

First Embodiment
Fig. 1 is a diagram showing an example of the overall configuration of an information processing system S according to the first embodiment. As shown in Fig. 1, the information processing system S includes a medical information processing device 10 and a CDSS (Clinical Decision Support System) 20. The medical information processing device 10 and the CDSS 20 are installed in a medical institution such as a hospital. The medical information processing device 10 is communicatively connected to the CDSS 20 via a network 300 such as an in-hospital LAN (Local Area Network). In addition, an electronic medical record system (not shown) or the like may be further connected to the medical information processing device 10 and the CDSS 20.

The CDSS20 provides information that is useful for medical professionals such as doctors to make clinical decisions. In this embodiment, the CDSS20 supports doctors by presenting a diagnosis result for a patient based on the results of learning data based on past performance through deep learning or other machine learning. More specifically, the CDSS20 of this embodiment outputs a treatment method that is estimated to be effective for the disease suffered by the patient, and the estimated treatment results when the treatment method is implemented.

CDSS20 includes a trained model that has been trained using training data. The trained model is, for example, a trained model that has been generated by deep learning such as a neural network. Deep learning techniques that can be used include, but are not limited to, decision trees, convolutional neural networks (CNNs), and recurrent neural networks (RNNs).

FIG. 2 is a diagram showing an example of a decision tree D used in the CDSS20 according to the first embodiment. As shown in FIG. 2, for example, the decision tree D estimates an outcome based on information about the patient's disease, such as the patient's medical images, blood test results, tumor markers, vital data, and blood test results such as the doctor's findings recorded in the electronic medical record. The outcome of the decision tree D shown in FIG. 2 is the presence or absence of side effects in the patient. The trained model used in the CDSSS20 may employ either a regression or classification method. The outcome output by the CDSSS20, i.e., the diagnosis result, may be not only a classification such as the presence or absence of side effects, but also a numerical value such as a side effect risk score estimated by a regression method.

CDSS20 is configured, for example, by a server device or a regular computer such as a PC (Personal Computer).

In addition, even after the start of operation, CDSS20 can improve its accuracy by learning new learning data. Improving accuracy means bringing the estimated treatment results output by CDSS20 closer to the actual treatment results of the patient. In this embodiment, the CDSS20 learning new learning data is referred to as updating CDSS20. The new learning data is an example of the first data in this embodiment.

In addition, the updated CDSS20 may output results that differ from those of the CDSS20 before the update. Ideally, updating the CDSS20 would improve the accuracy of the CDSS20 in all cases, but in reality, it may be difficult to improve diagnostic accuracy equally in all cases. Therefore, in this embodiment, the medical information processing device 10 described below prioritizes improving accuracy in cases in which the CDSS20 has a high degree of impact on the doctor's diagnosis.

Figure 3 is a diagram showing an example of the change in output before and after updating the CDSS 20 according to the first embodiment. In the example shown in Figure 3, the upper part shows the CDSS 20 before learning new learning data, and the lower part shows the CDSS 20 after learning the new learning data. Note that the "CDSS 20 before learning new learning data" may be a CDSS that has undergone multiple learning cycles in the past, or a CDSS that learned before the start of operation and has not yet performed any learning since the start of operation.

In the example shown in FIG. 3, the pre-update CDSS20 and the post-update CDSS20 output the diagnosis result for patient P when they receive input of patient information for patient P.

Specifically, CDSS20 outputs a treatment method estimated to be effective for the disease suffered by patient P, and the estimated treatment results when the treatment method is implemented. For example, CDSS20 outputs the type of drug to be administered and the dosage as the treatment method. CDSS20 also outputs a side effect risk score when the drug is administered as the treatment result. The side effect risk score is, for example, a numerical value that indicates the intensity of side effects caused by medication. Note that the information included in the diagnosis result is not limited to this.

The patient information includes information about the patient P's illness and attribute information about the patient P.

The attribute information of patient P includes, but is not limited to, the name, age, sex, height, and weight of patient P. Note that the ID of patient P may be used instead of the name.

The information regarding patient P's illness includes multiple types of test results regarding patient P. Multiple types of test results include, for example, vital data such as body temperature and blood pressure, blood test results such as white blood cell count, major markers, image measurement values such as tumor size, etc. Furthermore, in this embodiment, the test results include not only quantitative test results, but also findings by a doctor or nurse, such as the presence or absence of appetite or the presence or absence of loss of appetite. Note that findings may be included in the information regarding patient P's illness as information separate from the test results. Information regarding patient P's illness, such as the name of the illness, may also be included.

In the example shown in FIG. 3, the CDSS 20 before the update outputs a side effect risk score of "50" as a diagnostic result. The diagnostic result output by the CDSS 20 before the update is an example of a first diagnostic result in this embodiment.

In addition, in FIG. 3, not only CDSS 20 but also Doctor 4 diagnoses Patient P. Doctor 4's diagnosis result is a side effect risk score of "55." In the example shown in FIG. 3, Doctor 4 makes a diagnosis in consideration of the diagnosis result output from CDSS 20 before the update, his/her own experience, and other information. Doctor 4 is an example of a user according to this embodiment.

Even when using CDSS20, the final diagnosis is generally made by doctor 4. For this reason, this embodiment also assumes that for diagnosed patient P, a diagnosis by doctor 4 has already been issued.

The "true value" shown in Figure 3 indicates the actual treatment result of patient P. While the CDSS 20 and the diagnosis result of doctor 4 are estimated values before the actual treatment is administered, the "true value" is the treatment result determined by testing etc. after the treatment is actually administered. In this example, the actual severity of side effects of patient P is assumed to be about a side effect risk score of "58."

In the example shown in FIG. 3, even though the updated CDSS20 receives the same patient information as before the update, it outputs a different side effect risk score of "25" than before the update. The diagnosis result output by the updated CDSS20 is an example of the second diagnosis result in this embodiment.

The difference between the side effect risk score of "50" output by CDSS20 before the update and the true side effect risk score of "58" was "8". In contrast, the difference between the side effect risk score of "25" output by CDSS20 after the update and the true side effect risk score of "58" is "33". In this case, the difference between the diagnosis result of CDSS20 after the update and the true value is larger than the difference between the diagnosis result of CDSS20 before the update and the true value.

In such a case, after CDSS20 is updated, if Doctor 4 refers to the diagnosis results of CDSS20 when diagnosing a patient similar to Patient P, there is a possibility that the updated CDSS20 will lead Doctor 4's judgment to deviate from the true value.

The difference between the first diagnostic result output by the CDSS20 before the update and the true value is an example of a first difference in this embodiment. The difference between the second diagnostic result output by the CDSS20 after the update and the true value is an example of a second difference in this embodiment.

Returning to FIG. 1, the medical information processing device 10 performs an update simulation and processes the learning data before the CDSS 20 learns new learning data. The medical information processing device 10 is also referred to as, for example, a CDSS update support device.

The medical information processing device 10 is, for example, a server device or a PC, and includes a network interface 110, a memory circuit 120, an input interface 130, a display 140, and a processing circuit 150.

The NW interface 110 is connected to the processing circuit 150 and controls the transmission and communication of various data between the medical information processing device 10 and the CDSS 20. The NW interface 110 is realized by a network card, a network adapter, a NIC (Network Interface Controller), etc.

The memory circuitry 120 prestores various types of information to be used by the processing circuitry 150. The memory circuitry 120 also stores various programs.

The input interface 130 is realized by a trackball, a switch button, a mouse, a keyboard, a touchpad that performs input operations by touching the operation surface, a touch screen that integrates a display screen and a touchpad, a non-contact input circuit using an optical sensor, and a voice input circuit. The input interface 130 is connected to the processing circuit 150, and converts input operations received from an operator into electrical signals and outputs them to the processing circuit 150. Note that in this specification, the input interface is not limited to those that have physical operating parts such as a mouse and a keyboard. For example, an electrical signal processing circuit that receives electrical signals corresponding to input operations from an external input device provided separately from the device and outputs these electrical signals to the processing circuit 150 is also included as an example of the input interface 130.

The display 140 is a liquid crystal display, an organic electro-luminescence (OEL) display, or the like. The input interface 130 and the display 140 may be integrated. For example, the input interface 130 and the display 140 may be realized by a touch panel. The display 140 is an example of an output unit in this embodiment. In this embodiment, an example in which the display 140 is included in the medical information processing device 10 will be described, but the display 140 may be provided outside the medical information processing device 10.

The processing circuit 150 is a processor that reads out programs from the memory circuit 120 and executes them to realize functions corresponding to each program. The processing circuit 150 of this embodiment includes an acquisition function 151, a case extraction function 152, an influence variable extraction function 153, an output control function 154, a reception function 155, a data processing function 156, a generation function 157, a classification function 158, and a learning control function 159. The acquisition function 151 is an example of an acquisition unit. The case extraction function 152 is an example of a case extraction unit. The influence variable extraction function 153 is an example of an influence variable extraction unit. The output control function 154 is an example of an output control unit. The reception function 155 is an example of a reception unit. The data processing function 156 is an example of a data processing unit or a conversion unit. The generation function 157 is an example of a generation unit. The classification function 158 is an example of a classification unit. The learning control function 159 is an example of a learning control unit.

Here, for example, each processing function of the components of the processing circuit 150, namely, the acquisition function 151, the case extraction function 152, the influence variable extraction function 153, the output control function 154, the reception function 155, the data processing function 156, the generation function 157, the classification function 158, and the learning control function 159, is stored in the memory circuit 120 in the form of a program executable by a computer. The processing circuit 150 is a processor. For example, the processing circuit 150 realizes the function corresponding to each program by reading and executing the program from the memory circuit 120. In other words, the processing circuit 150 in the state in which each program has been read has each function shown in the processing circuit 150 of FIG. 1. In FIG. 1, the processing functions performed by the acquisition function 151, the case extraction function 152, the influence variable extraction function 153, the output control function 154, the reception function 155, the data processing function 156, the generation function 157, the classification function 158, and the learning control function 159 are described as being realized by a single processor, but it is also possible to configure the processing circuit 150 by combining multiple independent processors and having each processor execute a program to realize the function. Also, in FIG. 1, it is described that a single memory circuit 120 stores a program corresponding to each processing function, but it is also possible to configure multiple memory circuits to be distributed and have the processing circuit 150 read out the corresponding program from each memory circuit.

In the above description, an example was described in which the "processor" reads out and executes a program corresponding to each function from a storage circuit, but the embodiment is not limited to this. The term "processor" refers to circuits such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an Application Specific Integrated Circuit (ASIC), a programmable logic device (e.g., a Simple Programmable Logic Device (SPLD), a Complex Programmable Logic Device (CPLD), and a Field Programmable Gate Array (FPGA)). When the processor is, for example, a CPU, the processor realizes the function by reading out and executing a program stored in a storage circuit. On the other hand, when the processor is an ASIC, instead of storing a program in the storage circuit 120, the function is directly incorporated as a logic circuit in the circuit of the processor. Note that each processor in this embodiment is not limited to being configured as a single circuit for each processor, and may be configured as a single processor by combining multiple independent circuits to realize the function. Furthermore, multiple components in FIG. 1 may be integrated into a single processor to achieve the functions.

The acquisition function 151 acquires from the CDSS 20 the first diagnosis result, the second diagnosis result for each of the multiple diagnosed patients P, and the diagnosis result of the doctor 4 for each of the multiple diagnosed patients P under the support of the CDSS 20 before the update. The acquisition function 151 also acquires new learning data from the CDSS 20.

The new learning data is an example of the first data, and is learning data that is to be used to update CDSS20. In this case, the first diagnostic result is the diagnostic result by CDSS20 before learning the new learning data. Also, the second diagnostic result is the diagnostic result by CDSS20 after learning the new learning data.

Note that at the time of this acquisition, the update of CDSS20 has not been confirmed. For example, the second diagnosis result is not an actual update result, but a simulation result of what would happen if CDSS20 were updated. Note that the acquisition function 151 may acquire the second diagnosis result from an information processing device other than CDSS20, for example, an information processing device dedicated to simulation.

The new learning data in this embodiment includes multiple cases in which information about the disease for each of multiple diagnosed patients P is associated with a true value representing the treatment outcome for each of multiple diagnosed patients.

As described in FIG. 3, the information on the disease includes multiple types of test results for patient P. Note that cases do not necessarily have to be on a per-patient P basis; if one patient P visits the hospital multiple times for multiple diseases, each case may be included in the new learning data as a separate case. Also, as described above, the information on the disease includes attribute information for each of multiple diagnosed patients.

Figure 4 is a diagram showing an example of information acquired by the medical information processing device 10 according to the first embodiment. One record of the acquired data 800 shown in Figure 4 corresponds to one case.

As shown in FIG. 4, the patient information of each diagnosed patient P, the true value, the diagnosis result of the CDSS20 before the update (first diagnosis result), the diagnosis result of the CDSS20 after the update (second diagnosis result), and the diagnosis result of doctor 4 are associated for each case. Note that, for the sake of explanation, FIG. 4 shows the data in table format, but each piece of information included in the acquired data 800 only needs to be associated with each case, and does not need to be organized as a single table.

The acquisition function 151 sends the acquired new learning data, the first diagnosis result, the second diagnosis result, and the diagnosis result of the doctor 4 for each of the multiple diagnosed patients P with the support of the CDSS 20 before the update to the case extraction function 152.

Returning to FIG. 1, the case extraction function 152 extracts, from among the multiple cases included in the new learning data, cases of the multiple diagnosed patients P that are affected by the update of the CDSS 20 due to learning of the new learning data to a degree equal to or greater than a first specified standard, as presentation target cases, based on the first diagnosis result, the second diagnosis result, and the diagnosis result of the doctor 4 for each of the multiple diagnosed patients P with the support of the CDSS 20 before the update. The presentation target cases refer to cases that are displayed on the analysis screen described below. Note that in this embodiment, the case extraction function 152 extracts the presentation target case when it determines that one of the multiple cases is a presentation target case.

In this embodiment, a case in which the update of CDSS20 is affected by more than the first specified criterion refers to a case in which the second diagnostic result of CDSS20 after the update deviates from the true value more than the first diagnostic result of CDSS20 before the update.

In this embodiment, the difference between the first difference, which is the difference between the first diagnostic result and the true value, and the second difference, which is the difference between the second diagnostic result and the true value, is an example of an update impact index. More specifically, the update impact index is a value obtained by subtracting the first difference from the second difference. In this case, the larger the second difference is than the first difference, that is, the further the diagnostic result deviates from the true value due to the update, the larger the update impact index becomes.

In addition, in this embodiment, the case extraction function 152 further sets as a condition for a presented case that the influence on the diagnosis result of doctor 4 is greater than the third specified criterion. In this embodiment, the case extraction function 152 determines that the higher the similarity between the first diagnosis result and the diagnosis result of doctor 4 for each of the multiple diagnosed patients P under the support of CDSS20 before the update, the greater the influence on the diagnosis result of doctor 4. For this reason, the case extraction function 152 also sets as a condition for a presented case that the difference between the first diagnosis result and the diagnosis result of doctor 4 under the support of CDSS20 is equal to or less than the second specified criterion. Note that if the difference between the first diagnosis result and the diagnosis result of doctor 4 is equal to or less than the second specified criterion, the influence on the diagnosis result of doctor 4 is greater than the third specified criterion.

The difference between the first diagnosis result and the diagnosis result of doctor 4 with the support of CDSS20 is an example of a diagnostic influence index in this embodiment. The smaller the difference between the first diagnosis result and the diagnosis result of doctor 4 with the support of CDSS20, the larger the diagnostic influence index.

That is, the case extraction function 152 determines, among the multiple cases included in the new learning data, a case in which the first difference is equal to or less than the second difference and the difference between the first diagnosis result and the diagnosis result of doctor 4 with the support of CDSS20 is equal to or less than the second specified criterion as a presentation target case. The second specified criterion is, for example, a predetermined threshold value. More specifically, the case extraction function 152 extracts a case that satisfies both formulas (1) and (2) as a presentation target case.

Equation (1) specifies that the absolute value of the value obtained by subtracting the diagnosis result of Doctor 4 from the first diagnosis result of CDSS20 before the update is equal to or less than a threshold value. Furthermore, equation (2) specifies that the absolute value of the value obtained by subtracting the true value from the first diagnosis result of CDSS20 before the update is equal to or less than the absolute value of the value obtained by subtracting the true value from the second diagnosis result of CDSS20 after the update.

The threshold value in formula (1) may be set by a user such as doctor 4, or may be the average or median of the absolute values of the results of subtracting doctor 4's diagnosis result from the first diagnosis result of CDSS 20 before the update. Note that the extraction conditions for the cases to be presented are not limited to formulas (1) and (2).

In addition, when the diagnosis result is not expressed numerically, a condition other than the threshold value may be used as the second prescribed criterion. For example, when the diagnosis result is output as a binary value of "yes" or "no", such as the presence or absence of side effects, the second prescribed criterion may be "there is no difference between the first diagnosis result and the diagnosis result of doctor 4 under the support of CDSS20". In this case, if there is no difference between the first diagnosis result and the diagnosis result of doctor 4 under the support of CDSS20, it is equivalent to the second prescribed criterion. In such a case, the "difference is equal to or less than the second prescribed criterion", so the condition that "the difference between the first diagnosis result and the diagnosis result of doctor 4 under the support of CDSS20 is equal to or less than the prescribed criterion" is satisfied. Alternatively, the case extraction function 152 may convert the non-numerical diagnosis result into a numerical value and then calculate the difference.

In other words, a patient P who satisfies formula (2) is a patient P whose diagnostic result accuracy has decreased before and after updating the CDSS20. For such a patient P, the diagnostic result obtained by the CDSS20 is significantly affected by the update.

As illustrated in formula (1), a patient P for whom the difference between the first diagnosis result and the diagnosis result of doctor 4 with the support of CDSS20 is equal to or less than the second prescribed standard is, in other words, a patient P for whom the diagnosis result of doctor 4 is similar to the first diagnosis result. In this embodiment, when the first diagnosis result and the diagnosis result of doctor 4 with the support of CDSS20 are similar, it is assumed that doctor 4 may have been influenced by the diagnosis result of CDSS20 in diagnosing the patient P.

Here, the effect of the CDSS20 on the diagnosis results of doctor 4 will be described with reference to Figures 5 to 7. Note that diagnosed patients P1 to P3 shown in Figures 5 to 7 are different individuals. Note that when there is no particular distinction between the individual diagnosed patients P1 to P3, they will simply be referred to as patient P.

Figure 5 is a diagram showing an example of the relationship between the diagnosis results of the CDSS 20 and the diagnosis results of doctor 4 according to the first embodiment.

In the example shown in FIG. 5, the CDSS20 before the update output a first diagnosis result indicating that patient P1 will have side effects. Similarly to the CDSS20 before the update, doctor 4 also outputs a diagnosis result indicating that patient P1 will have side effects. If the second prescribed criterion is that "the first diagnosis result and doctor 4's diagnosis result are identical," in other words, if the difference threshold is "0," then the difference between the first diagnosis result for patient P1 and doctor 4's diagnosis result with the support of CDSS20 is equal to or less than the second prescribed criterion.

In this case, it is highly likely that Doctor 4 made a diagnosis based on the first diagnostic result of CDSS20 before the update. For example, if another patient P with similar test results to those of patient P1 is received, Doctor 4 may make a diagnosis under the influence of the diagnostic result of CDSS20, as in the case of patient P1.

The treatment result for patient P1, that is, the true value, was "side effects." In this case, the first diagnosis result is equal to the true value. In contrast, the second diagnosis result of the updated CDSS20 is "no side effects," which is different from the true value. In this case, the first difference is smaller than the second difference, and equation (2) is satisfied.

For patient P1 shown in FIG. 5, the first difference is equal to or less than the second difference, and the difference between the first diagnosis result and the diagnosis result of doctor 4 with the support of CDSS20 is equal to or less than the second prescribed standard, so the case extraction function 152 extracts the case of patient P1 as a case to be presented.

Also, FIG. 6 is a diagram showing another example of the relationship between the diagnosis result of the CDSS20 according to the first embodiment and the diagnosis result of doctor 4. In the example shown in FIG. 6, doctor 4 issues a diagnosis result different from the first diagnosis result of the CDSS20 before the update. If the second prescribed criterion is that "the first diagnosis result and the diagnosis result of doctor 4 are identical," that is, if the difference threshold is "0," the case of patient P2 does not satisfy the criterion. As shown in FIG. 6, if the difference between the diagnosis result of the CDSS20 before the update and the diagnosis result of doctor 4 is large, it is generally estimated that the impact of the CDSS20 before the update on doctor 4 is small. In this case, the case extraction function 152 does not extract the case of patient P2 as a case to be presented.

Also, FIG. 7 is a diagram showing yet another example of the relationship between the diagnosis result of CDSS20 according to the first embodiment and the diagnosis result of doctor 4. In the example shown in FIG. 7, the first diagnosis result of CDSS20 before the update for patient P3 is "side effects present", the second diagnosis result of CDSS20 after the update is "no side effects", and the true value is "no side effects". In other words, in the example shown in FIG. 7, the accuracy of the diagnosis of patient P3 by CDSS20 has improved due to the update. In this case, the first difference is larger than the second difference, and formula (2) is not satisfied. In this case, the case extraction function 152 does not extract the case of patient P3 as a case to be presented.

Note that although Figures 5 to 7 show examples in which the same doctor 4 makes a diagnosis, different doctors 4 may make the diagnosis.

The case extraction function 152 sends to the output control function 154 those corresponding to the extracted presentation target cases among the new learning data, the first diagnosis result, the second diagnosis result, and the diagnosis results of the doctor 4 for each of the multiple diagnosed patients P under the support of the CDSS 20 before the update. The case extraction function 152 also sends to the output control function 154 the diagnosis impact index and the update impact index corresponding to the extracted presentation target cases.

Returning to FIG. 1, the influence variable extraction function 153 extracts, from among the multiple types of test results for each of the multiple diagnosed patients P included in the new learning data, items that influenced the change in the diagnosis results of the CDSS 20 before and after the update as influence variables. Note that in this embodiment, since the influence variables are determined by the classification function 158 described below, more precisely, the influence variable extraction function 153 extracts "influence variable candidates." Hereinafter, the influence variable candidates will also be referred to as influence variables unless a special distinction is made.

In this embodiment, each item included in the test result is called a variable. In other words, when the first diagnostic result and the second diagnostic result of CDSS20 differ, the influencing variable extraction function 153 estimates which variable is responsible for the change. The influencing variable extraction function 153 extracts the variables that influenced the change in the diagnostic result of CDSS20 as influencing variables.

Figure 8 is a diagram showing an example of a method for extracting influential variables according to the first embodiment. As shown in Figure 8, the influential variable extraction function 153 inputs, into the prediction model 70, patient information for each case included in the new learning data and whether or not the diagnosis results of the CDSS 20 have changed before and after the update for each case.

The presence or absence of a change is also referred to as a label for the patient information. The influencing variable extraction function 153 associates the label "changed" with cases in which there was a change in the diagnostic result of CDSS20 before and after the update, and "unchanged" with cases in which there was no change in the diagnostic result of CDSS20 before and after the update, and inputs these into the prediction model 70. Cases in which there was a change in the diagnostic result of CDSS20 before and after the update are also referred to as "changed cases", and cases in which there was no change in the diagnostic result of CDSS20 before and after the update are also referred to as "unchanged cases".

More specifically, whether or not the diagnostic result of CDSS20 has changed before and after the update refers to whether or not there is a difference between the first diagnostic result and the second diagnostic result. When the first diagnostic result and the second diagnostic result are expressed numerically, the influence variable extraction function 153 may determine that the difference between the first diagnostic result and the second diagnostic result is "unchanged" if it is equal to or less than a specified threshold, and that the difference is "changed" if it is greater than the specified threshold.

As described above, the test results include, for example, vital data such as patient P's body temperature and blood pressure, blood test results such as white blood cell count, image measurements such as major markers and tumor size, and findings by a doctor or nurse. The prediction model 70 outputs the magnitude of influence that each item of these test results, i.e., each variable, has on the change in the diagnosis result of the CDSS 20 before and after the update. In this embodiment, the magnitude of influence that each variable has on the change in the diagnosis result of the CDSS 20 before and after the update is called the importance of each variable. In the example shown in FIG. 8, the prediction model 70 expresses the importance of each variable as a decimal value between 0 and 1. The larger the value, the greater the influence that the variable has on the change in the diagnosis result of the CDSS 20 before and after the update.

The prediction model 70 is, for example, a trained model generated by machine learning such as random forest. The machine learning method is not limited to this, and other known machine learning methods can be adopted. The prediction model 70 may be stored in the memory circuit 120, or may be incorporated in the influential variable extraction function 153.

In this embodiment, the influencing variable extraction function 153 determines that a variable whose importance is greater than or equal to a specified criterion is an influencing variable. The influencing variable extraction function 153 extracts an influencing variable when it determines that a variable from among multiple variables is an influencing variable.

The influence variable extraction function 153 sends the extracted influence variables and the importance of the influence variables to the output control function 154.

Returning to FIG. 1, the output control function 154 causes the presentation target cases extracted by the case extraction function 152 to be output to the display 140 in association with the diagnostic impact index. In this embodiment, the screen on which the presentation target cases and the diagnostic impact index are displayed is called an analysis screen. In this embodiment, the term "output" includes "display." Furthermore, the output method is not limited to display. For example, the output control function 154 may output the presentation target cases to an information processing device other than the medical information processing device 10 in association with the diagnostic impact index.

Figure 9 is a diagram showing an example of an analysis screen according to the first embodiment. As shown in Figure 9, the output control function 154 displays a list 91 of influencing variables extracted by the influencing variable extraction function 153 on the analysis screen on the display 140. The list 91 includes, as an example, the name and importance of the variable. The user can select any influencing variable from the list 91 as the subject of analysis. In the example shown in Figure 9, three influencing variables, "white blood cell count," "tumor marker," and "body temperature," have been selected by the user.

The output control function 154 causes the examples to be presented to be output to the display 140 in association with the influencing variables. More specifically, the output control function 154 causes the analysis screen to display a graph with two of the influencing variables selected by the user as axes. In the example shown in FIG. 9, three types of graphs are displayed: a graph with "tumor marker" and "white blood cell count" as axes, a graph with "tumor marker" and "body temperature" as axes, and a graph with "white blood cell count" and "body temperature" as axes. Note that in the example shown in FIG. 9, only the influencing variables selected by the user are used as the axes of the graphs, but all of the extracted influencing variables may be used as the axes of the graphs.

On each graph, the presentation target cases extracted by the case extraction function 152 are mapped. For example, point 8 on the graph with the axes of "tumor marker" and "white blood cell count" is a case of patient "XX" whose "white blood cell count" is "2500/ ^mm3 " and whose "tumor marker" value is "8 ng/ml." The graph shown in Fig. 9 is an example of a plane defined by the values of two types of test results.

In addition, when a user selects a point representing a case, for example with a mouse pointer, the output control function 154 may display the patient information, true value, first diagnosis result, second diagnosis result, and doctor 4's diagnosis result corresponding to the case.

The output control function 154 also associates the diagnosis impact index and the update impact index with the presentation target cases and outputs them to the display 140. In the example shown in FIG. 9, the output control function 154 highlights on the graph the presentation target cases whose diagnosis impact index or update impact index is equal to or greater than the threshold. This allows the user to know whether the diagnosis impact index or update impact index of each presentation target case mapped on the graph is equal to or greater than the threshold. In the example shown in FIG. 9, points indicating presentation target cases whose diagnosis impact index or update impact index is equal to or greater than the threshold are displayed larger than points indicating presentation target cases whose diagnosis impact index or update impact index is smaller than the threshold.

The threshold value of the diagnostic impact index and the threshold value of the update impact index can each be changed by the user in the threshold adjustment section 90 on the analysis screen. The output control function 154 displays the threshold adjustment section 90 including operable knobs and bars. In FIG. 9, the diagnostic impact index is represented as "the difference between the CDSS (before update) and the user's diagnosis result," and the update impact index is represented as "the difference between the CDSS before and after the update and the true value," so that the user can understand the content of each index.

In the example shown in FIG. 9, the cases to be highlighted are those for which both the diagnostic impact index and the update impact index are above a threshold value, but the condition for highlighting may also be that at least one of the diagnostic impact index and the update impact index is above a threshold value.

In the example shown in Figure 9, the case of patient name "XX" corresponding to point 8 has both a diagnosis impact index and an update impact index that are greater than or equal to the threshold. In other words, for the case of patient name "XX", although the impact on doctor 4's diagnosis is greater than or equal to the threshold, the update has caused the diagnosis result to move further from the true value by more than the threshold than before the update.

Presentation target cases that correspond to a diagnostic impact index equal to or greater than a threshold value set by the user and an update impact index equal to or greater than the threshold value are given priority in improving the diagnostic accuracy of CDSS20 in the update process described below.

Returning to FIG. 1, the reception function 155 receives operations by the user via the input interface 130. For example, the reception function 155 receives a user operation to adjust the threshold of the diagnostic impact index or the threshold of the update impact index in the threshold adjustment section 90 on the analysis screen. The reception function 155 also receives an operation in which the user selects a variable from the list of influencing variables 91 on the analysis screen.

The reception function 155 sends the received operation content to the output control function 154 and the data processing function 156.

The data processing function 156 processes new learning data generated by the user based on the thresholds for the diagnostic impact index and the update impact index determined by the user, and the importance of each variable calculated by the impact variable extraction function 153.

More specifically, the data processing function 156 includes a generation function 157 and a classification function 158.

The generation function 157 copies cases whose diagnostic impact index is equal to or greater than a threshold value from among the multiple cases included in the new learning data to generate new cases, and adds the new cases to the new learning data.

Figure 10 is a diagram showing an example of adding a new case according to the first embodiment. As shown in Figure 10, the generation function 157 duplicates the case of patient name "XX" among the multiple cases included in the new learning data, in which both the diagnosis impact index and the update impact index are equal to or greater than the threshold value, to generate cases of patient name "XX2" and patient name "XX3". The cases of patient name "XX2" and patient name "XX3" are pseudo data that are not actual diagnosis histories. The cases of patient name "XX2" and patient name "XX3" contain the same information as the case of patient name "XX" except for identification information such as the patient name.

By duplicating in this way, the number of cases in the new learning data that have similar patient information and true values to the case with patient name "XX" increases, so that when updating the CDSS20, learning is performed to improve the diagnostic accuracy for cases with patient information similar to the patient name "XX". The generation function 157 duplicating cases to increase the number of samples in the new learning data that have similar patient information and true values is an example of a method of increasing the weight for the case.

The generation function 157 may automatically select cases to be copied, in addition to cases identified by the diagnostic impact index threshold and the update impact index threshold set by the user.

The classification function 158 also classifies multiple variables contained in the new learning data into "influence variables" that have a large impact on the update of CDSS20 and "non-influence variables" that have a small impact on CDSS20.

Figure 11 is a diagram showing an example of variable classification according to the first embodiment. In the example shown in Figure 11, among the influencing variable candidates extracted by the influencing variable extraction function 153, three variables selected by the user on the analysis screen are classified as "influencing variables", and the other influencing variable candidates are classified as "non-influencing variables". In addition, variables that were not extracted as influencing variable candidates are also classified as "non-influencing variables".

The classification function 158 may classify the influencing variables into "influencing variables" and "non-influencing variables" based on the importance of each influencing variable candidate, rather than on the basis of user selection. Alternatively, the classification function 158 may classify all influencing variable candidates extracted by the influencing variable extraction function 153 into "influencing variables", and classify only variables that were not extracted as influencing variable candidates into "non-influencing variables".

Returning to FIG. 1, the learning control function 159 controls the learning process in learning the CDSS 20 so that, among the multiple cases included in the new learning data, the case in which the difference between the first diagnosis result and the diagnosis result of doctor 4 is smaller is given a higher priority.

More specifically, the learning control function 159 controls the learning process in learning CDSS20 so as to minimize the difference between the second diagnosis result of CDSS20 and the true value.

Furthermore, the learning control function 159 classifies the multiple types of test results included in the disease information into first test results that have a large influence on the changes between the first and second diagnostic results, and second test results that have a smaller influence on the changes between the first and second diagnostic results than the first test results. The learning control function 159 also inputs the first test results into the first model, and inputs the second test results into the second model. Then, the learning control function 159 adjusts the weighting coefficient so that the weight of the output result of the first model is large, and then calculates the weighted average of the output result of the first model and the output result of the second model, and outputs the calculation result of the weighted average as a prediction result. In other words, the learning control function 159 adjusts the influence of the "influence variable" on the update by adjusting the weighting coefficient. It is assumed that both the first model and the second model are CDSS.

Figure 12 is a diagram showing an example of the learning process according to the first embodiment. The learning control function 159 updates the CDSS 20 based on the processed learning data obtained by processing new learning data by the data processing function 156. Note that the "correct label" shown in Figure 12 refers to the true value.

As shown in FIG. 12, the learning control function 159 obtains a first learning result 61 in which only the "influence variables" of the new learning data have been learned, and a second learning result 62 in which only the "non-influence variables" of the new learning data have been learned. The learning control function 159 then multiplies the second learning result 62 by a weighting coefficient, and uses the result as further learning data together with the first learning result 61 not multiplied by a weighting coefficient, thereby increasing the influence of the "non-influence variables" in updating the CDSS 20 and decreasing the influence of the "influence variables". Note that the values of the weighting coefficients shown in FIG. 12 are merely examples and are not limited to these.

As shown in FIG. 12, new cases copied by the generation function 157 are added to the processed learning data. Therefore, the learning control function 159 performs learning of the CDSS 20 by applying two types of weighting: a first weighting due to the addition of new cases, and a second weighting for the "non-influencing variables."

The learning control function 159 also applies weighting to the learning process algorithm. This weighting is such that, among the multiple cases included in the processed learning data, the case in which the difference between the first diagnosis result and the diagnosis result of doctor 4 is small, that is, the case in which the value of the diagnostic impact index is large, is given a higher priority.

In this embodiment, the learning control function 159 performs learning using a minimization model of a weighted loss function. For example, the learning control function 159 employs the Mean Square Error (MSE). Here, a model for generally bringing the second diagnosis result of the CDSS 20 closer to the true value using MSE is shown as equation (3).

However, the learning control function 159 of this embodiment performs learning using a minimization model of equation (4) which is a weighted version of equation (3). In equation (4), the weighting coefficient is increased for cases in which the difference between the first diagnosis result and the diagnosis result of doctor 4 is small. For this reason, equation (4) not only brings the second diagnosis result closer to the true value, but also increases the influence of cases with large values of the diagnostic influence index on the learning of CDSS20. Note that the learning method is not limited to MSE, and other machine learning methods may be adopted.

In this embodiment, the learning control function 159 of the medical information processing device 10 controls the learning process of the CDSS 20, but the learning control function 159 may also be a function of the CDSS 20.

Next, we will explain the flow of the update support process for the CSDD 20 executed by the medical information processing device 10 configured as described above.

Figure 13 is a flowchart showing an example of the flow of update support processing according to the first embodiment.

First, the acquisition function 151 acquires data to be used in the update support process from the CDSS 20 (S1). Specifically, the acquisition function 151 acquires the first diagnosis result, the second diagnosis result for each of the multiple diagnosed patients P, the diagnosis result of the doctor 4 for each of the multiple diagnosed patients P under the support of the CDSS 20 before the update, and new learning data.

Next, the case extraction function 152 extracts, from among the multiple cases included in the new learning data, cases of the multiple diagnosed patients P that are affected by the update of the CDSS 20 due to learning of the new learning data by a specified standard or more, as cases to be presented, based on the first diagnosis result, the second diagnosis result, and the diagnosis result of the doctor 4 for each of the multiple diagnosed patients P with the support of the CDSS 20 before the update (S2). The case extraction function 152 also calculates an update impact index and a diagnosis impact index for each case.

The influencing variable extraction function 153 also extracts, as influencing variables, items that influenced the changes in the diagnostic results of the CDSS20 before and after the update from among the multiple types of test results for each of the multiple diagnosed patients P included in the new learning data (S3).

Then, the output control function 154 causes the display 140 to display an analysis screen that shows the presentation target cases extracted by the case extraction function 152 in association with the update impact index, the diagnostic impact index, and the impact variables extracted by the impact variable extraction function 153 (S4). In addition, the reception function 155 receives various operations by the user on the analysis screen.

The generation function 157 and classification function 158 included in the data processing function 156 process new learning data based on the user's setting of the threshold for the update impact index, the setting of the threshold for the diagnosis impact index, and the selection of the impact variable (S5). Note that the generation function 157 and classification function 158 may process new learning data based on a specified standard or a calculation result, etc., rather than a user operation.

Then, the learning control function 159 uses the processed learning data to update the CDSS 20 (S6). At this point, the processing of this flowchart ends.

In this way, the medical information processing device 10 of this embodiment displays on the display 140, among the cases included in the new learning data, cases that are affected by the update of the CDSS20 by a first specified criterion or more, in association with a diagnostic impact index. Therefore, the user can easily grasp, among the presented cases that are affected by the update of the CDSS20 by a specified criterion or more, those that have a large impact on the diagnosis of the doctor 4, and the user can select cases that should be prioritized in learning the CDSS20. Therefore, according to the medical information processing device 10 of this embodiment, it is possible to support preferentially improving the diagnostic accuracy of cases that are important when the doctor 4 makes a diagnosis in learning the CDSS20.

The medical information processing device 10 of this embodiment also associates an update impact index with the presentation target cases and displays it on the display 140. Therefore, according to the medical information processing device 10 of this embodiment, the user can easily grasp which of the presentation target cases are strongly affected by the update of the CDSS 20.

The medical information processing device 10 of this embodiment also extracts, from among the multiple cases included in the new learning data, cases in which the first difference is smaller than the second difference and the difference between the first diagnosis result and the diagnosis result of doctor 4 with the assistance of CDSS20 is equal to or less than a second specified standard, as examples to be presented. Therefore, according to the medical information processing device 10 of this embodiment, the user can easily grasp cases in which the diagnosis result of CDSS20 was influenced when doctor 4 made a diagnosis.

The diagnostic influence index of this embodiment indicates the degree of influence that CDSS20 has on the diagnosis of doctor 4. The diagnostic influence index has a larger value as the difference between the first diagnosis result and the diagnosis result of the doctor with the support of CDSS20 becomes smaller. Therefore, the medical information processing device 10 of this embodiment can easily determine whether or not a case is one in which the diagnosis result of CDSS20 influenced doctor 4's diagnosis.

The medical information processing device 10 of this embodiment also copies cases whose diagnostic influence index is equal to or greater than a threshold among the multiple cases included in the new learning data to generate new cases, and adds the new cases to the new learning data. Therefore, according to the medical information processing device 10 of this embodiment, by increasing the number of samples in the new learning data whose diagnostic influence index is equal to or greater than a threshold, it is possible to focus on improving the diagnostic accuracy of the CDSS 20 for cases that are likely to be referred to by the doctor 4 when making a diagnosis.

The medical information processing device 10 of this embodiment also controls the learning process in learning the CDSS20 so that, among the multiple cases included in the new learning data, the case in which the difference between the first diagnosis result and the diagnosis result of the doctor 4 under the support of the CDSS20 is smaller is given a higher priority. Therefore, according to the medical information processing device 10 of this embodiment, it is possible to focus on improving the diagnostic accuracy of the CDSS20 for cases that are likely to be referred to by the doctor 4 when making a diagnosis.

The medical information processing device 10 of this embodiment also controls the learning process in learning the CDSS20 so as to minimize the difference between the second diagnosis result of the CDSS20 and the true value. Therefore, according to the medical information processing device 10 of this embodiment, the accuracy of the diagnosis of the CDSS20 can be improved by learning.

In addition, the medical information processing device 10 of this embodiment controls the learning process so as to lower the priority of the multiple types of test results included in the new learning data that affect changes in the first and second diagnostic results. Therefore, the medical information processing device 10 of this embodiment reduces significant changes in the diagnostic results due to updates to the CDSS 20.

The medical information processing device 10 of this embodiment also extracts items related to the difference between the first and second diagnostic results from among the items of multiple types of test results included in the new learning data as influencing variables that affect the update of the CDSS 20, and displays the cases to be presented on the display 140 in association with the influencing variables. Therefore, according to the medical information processing device 10 of this embodiment, the items related to the difference between the first and second diagnostic results can be used as the analysis axis of the cases to be presented.

In this embodiment, the influencing variable extraction function 153 and the classification function 158 are described as different functions, but the influencing variable extraction function 153 and the classification function 158 may be configured as one function. The influencing variable extraction function 153 extracts, for example, test items with high importance as influencing variables with high importance for the update of the CDSS20. The influencing variable extraction function 153 extracts influencing variables with high importance based on the variable importance output from a prediction model that has been trained using the label indicating the classification of "important" or "unimportant" for each test item as the correct answer. In addition, cases in which the diagnosis result by the CDSS20 before the update is close to the diagnosis result by the doctor 4 (high diagnostic influence) and the difference between the diagnosis result by the CDSS20 before the update and the diagnosis result by the CDSS20 after the update is large (high update influence) are considered important, and cases that are not important are considered unimportant. More specifically, this is a test item for cases in which the difference between the diagnosis result by the CDSS20 before the update and the diagnosis result by doctor 4 with the support of the CDSS20 is equal to or less than the second prescribed criterion, and the difference between the diagnosis result by the CDSS20 before the update and the diagnosis result by the CDSS20 after the update is equal to or more than the fourth prescribed criterion. The fourth prescribed criterion may be a numerical value or another condition.

Second Embodiment
In the first embodiment described above, the medical information processing device 10 displays, on the analysis screen, cases related to diagnosed patients P. In the second embodiment, the medical information processing device 10 also displays, on the analysis screen, cases of patients who have not yet been diagnosed.

The information processing system S of this embodiment includes a medical information processing device 10 and a CDSS 20, similar to the first embodiment.

Furthermore, the medical information processing device 10 of this embodiment includes a network interface 110, a memory circuitry 120, an input interface 130, a display 140, and a processing circuitry 150, similar to the first embodiment.

The processing circuit 150 of this embodiment, like the first embodiment, has an acquisition function 151, a case extraction function 152, an influence variable extraction function 153, an output control function 154, a reception function 155, a data processing function 156, a generation function 157, a classification function 158, and a learning control function 159. The case extraction function 152, the influence variable extraction function 153, the reception function 155, the data processing function 156, the generation function 157, the classification function 158, and the learning control function 159 have the same functions as the first embodiment.

In addition to the functions of the first embodiment, the acquisition function 151 of this embodiment acquires undiagnosed patient information including multiple undiagnosed cases related to information on diseases for multiple patients for which a diagnosis has not been made by the doctor 4 and the CDSS 20. The source of the undiagnosed patient information is, for example, an electronic medical record system.

The undiagnosed patient information is an example of the second data in this embodiment. The undiagnosed patient information includes information about diseases for multiple patients for which a diagnosis has not been performed. In the undiagnosed patient information, for example, information about a disease associated with one patient is considered to be one undiagnosed case. The undiagnosed patient information includes multiple undiagnosed cases.

In addition, the disease-related information contained in the undiagnosed patient information includes multiple types of test results, i.e., variables, for multiple patients who have not been diagnosed.

In addition to the functions of the first embodiment, the output control function 154 of this embodiment causes the display 140 to display information indicating the likelihood that the diagnosis results of the CDSS 20 for multiple patients included in the undiagnosed patient information will change before and after learning new learning data.

Information indicating the likelihood that the diagnostic results of CDSS20 for multiple patients included in the undiagnosed patient information will change before and after learning new learning data is, for example, information indicating whether multiple undiagnosed cases included in the undiagnosed patient information fall into a half-value variable for which the difference in the diagnostic results of CDSS20 before and after the update is large.

14 is a diagram showing an example of an analysis screen according to the second embodiment. In the example shown in FIG. 14, the output control function 154 displays on the graph an area 93c-93c (hereinafter simply referred to as area 93) where the difference between the diagnostic results of the CDSS 20 before and after the update of the CDSS 20 is large. The difference between the diagnostic results of the CDSS 20 before and after the update of the CDSS 20 is large means that the difference between the first diagnostic result and the second diagnostic result is equal to or greater than the fourth prescribed criterion. As described above, the fourth prescribed criterion may be a numerical value or another condition. In the example shown in FIG. 14, the area 93 is the range on the graph with the variable "tumor marker value" on the vertical axis and the variable "body temperature" on the horizontal axis where the difference between the first diagnostic result and the second diagnostic result of the diagnosed patient P is equal to or greater than the fourth prescribed criterion.

The output control function 154 maps and displays, on a graph defined by two types of variables, multiple diagnosed patient P cases included in the new learning data and multiple undiagnosed cases related to multiple patients who have not been diagnosed, included in the undiagnosed patient information, as shown in the graph in FIG. 14. The graph on the analysis screen shown in FIG. 14 is an example of a plane defined by the values of two types of test results.

For this reason, if an undiagnosed case mapped on the graph falls within region 93, there is a high possibility that the diagnosis result of CDSS20 for the patient of that medical diagnosis case will change before and after learning new training data. If many undiagnosed cases fall within region 93, it is advisable for the user to carefully select the threshold value, etc., so that updating CDSS20 does not reduce the accuracy of diagnosis for patients corresponding to these undiagnosed cases.

In this way, the medical information processing device 10 of this embodiment acquires undiagnosed patient information, and displays on the display 140 information indicating the likelihood that the diagnosis results of the CDSS20 for multiple patients included in the undiagnosed patient information will change before and after learning new learning data. Therefore, according to the medical information processing device 10 of this embodiment, in addition to providing the effects of the first embodiment, the user can easily understand the impact that an update to the CDSS20 will have on diagnoses for patients who have not yet been diagnosed. For example, by using the medical information processing device 10 of this embodiment, the user can understand, before updating the CDSS20, the extent to which the update will affect the diagnoses of patients who are scheduled to be diagnosed in the future.

The medical information processing device 10 of this embodiment also maps and displays on a graph defined by two types of variables multiple diagnosed patient P cases included in the new learning data and multiple undiagnosed cases related to multiple patients who have not been diagnosed and included in the undiagnosed patient information. Therefore, the medical information processing device 10 of this embodiment allows the user to visually grasp the distribution of diagnosed patient cases and the distribution of undiagnosed patient cases.

(Variation 1)
Although the analysis screen in each of the above-described embodiments displays the simulation results when the CDSS 20 is updated under one condition, simulation results under a plurality of different conditions may be displayed on one screen.

Figure 15 is a diagram showing an example of an analysis screen according to the first modified example. As shown in Figure 15, the output control function 154 of this modified example outputs, on the same screen, the examples to be presented when the CDSS 20 is updated under a plurality of different conditions, for each of the plurality of different conditions.

The first screen area 7a and the second screen area 7b of the analysis screen shown in FIG. 15 each represent a presentation target case in CDSS 20 that has been updated under different conditions. The content displayed in each screen area is the same as that of the analysis screen in the first embodiment described in FIG. 9 or the analysis screen in the second embodiment described in FIG. 14. For example, the first screen area 7a and the second screen area 7b include threshold adjustment sections 90a, 90b, lists of influencing variables 91a, 91b, and a graph with the selected variable as the axis.

Different conditions include, but are not limited to, for example, that the content of the new learning data is different, that the algorithm of the CDSS20 to be updated is different, or that the past learning data used in the update prior to the update with the new learning data is different.

In this way, simulation results under multiple different conditions are displayed on one screen, allowing users to easily compare and consider which conditions they should adopt.

(Variation 2)
In the analysis screen of each of the above-described embodiments, the diagnostic influence index is determined based on the difference between the first diagnostic result and the diagnostic result of the doctor 4 with the support of the CDSS 20. However, the method of calculating the magnitude of the influence of the CDSS 20 on the diagnosis of the doctor 4 is not limited to this.

For example, the degree of influence of CDSS20 on doctor 4's diagnosis may be self-reported by doctor 4. The diagnostic influence index in this modification is the report of doctor 4 who diagnosed patient P of the presented case, regarding the degree of influence of CDSS20 at the time of diagnosis. The report result is the result of doctor 4 himself reporting whether or not doctor 4 was influenced by the diagnosis result of CDSS20 at the time of diagnosis. Note that the report result may be information that expresses not only the presence or absence of influence, but also the magnitude of the influence as a numerical value or level. The acquisition function 151 in this modification acquires the report result for each case from, for example, an electronic medical record system. Alternatively, the reception function 155 may receive the input of the report result for each case by doctor 4.

(Variation 3)
The configuration of the information processing system S is not limited to the above-mentioned example. For example, in each of the above-mentioned embodiments, the CDSS 20 and the medical information processing device 10 are described as separate devices, but the CDSS 20 and the medical information processing device 10 may be integrated into one device. In this case, the CDSS 20 may be an example of the medical information processing device.

The information processing system S may also include multiple CDSSs 20. Furthermore, some or all of the CDSSs 20 and the medical information processing device 10 may be realized in a cloud environment.

(Variation 4)
In addition, the trained models in each of the above-mentioned embodiments include "self-learning models" that further update the internal algorithms of the trained models by allowing a user to provide feedback on the results output by these trained models.

Furthermore, in each of the above-mentioned embodiments, the processing described as an example of a method for realizing a trained model may be realized by a method other than machine learning or deep learning.

(Variation 5)
In each of the above-described embodiments, the user adjusts the threshold of the diagnosis influence index or the threshold of the update influence index in the threshold adjustment unit 90 on the analysis screen, thereby indirectly selecting the case to be prioritized in updating the CDSS 20. The selection method is not limited to this, and the user may directly select the case to be prioritized in updating the CDSS 20 by selecting the point 8 displayed on the graph on the analysis screen.

According to at least one of the embodiments described above, it is possible to assist in the training of the clinical decision support system in preferentially improving the diagnostic accuracy of cases that are important when a doctor makes a diagnosis.

Although several embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations of embodiments can be made without departing from the spirit of the invention. These embodiments and their modifications are within the scope of the invention and its equivalents as set forth in the claims, as well as the scope and spirit of the invention.

4 Doctor 7a First screen area 7b Second screen area 8 Point 10 Medical information processing device 20 CDSS
61, 62 Learning result 70 Prediction model 90a, 90b Threshold adjustment unit 91a, 91b List 93, 93a to 93c Area 110 NW interface 120 Memory circuit 130 Input interface 140 Display 150 Processing circuit 151 Acquisition function 152 Case extraction function 153 Influence variable extraction function 154 Output control function 155 Reception function 156 Data processing function 157 Generation function 158 Classification function 159 Learning control function 300 Network 800 Acquired data P, P1 to P3 Patient S Information processing system

Claims (15)

an acquisition unit that acquires a first diagnosis result for each of a plurality of diagnosed patients by the clinical decision support system before learning first data, a second diagnosis result for each of the plurality of diagnosed patients by the clinical decision support system after learning the first data, a doctor's diagnosis result for each of the plurality of diagnosed patients, and the first data;
a case extraction unit that calculates an update influence index and a diagnosis influence index for each of a plurality of cases included in the first data based on the first diagnosis result, the second diagnosis result, and the doctor's diagnosis result, and extracts , from the plurality of cases included in the first data, cases in which the update influence index is equal to or greater than a first specified standard and the diagnosis influence index is equal to or less than a second specified standard , as presentation target cases;
an output control unit that causes an output unit to output the presented case examples in association with a diagnosis influence index indicating the degree of influence that the clinical decision support system has on a doctor's diagnosis in the presented case examples, and the update influence index ;
The first data includes a plurality of cases in which information on a disease for each of the plurality of diagnosed patients is associated with a true value representing a treatment outcome for each of the plurality of diagnosed patients;
the update influence index is an index based on a difference between the first diagnosis result and the true value and a difference between the second diagnosis result and the true value,
The diagnostic influence index is an index based on a difference between the first diagnostic result and the doctor's diagnostic result.
Medical information processing equipment. the update influence index represents a magnitude of a difference between a first difference, which is a difference between the first diagnosis result and the true value, and a second difference, which is a difference between the second diagnosis result and the true value ;
The medical information processing device according to claim 1 . the case extraction unit extracts, from among the multiple cases included in the first data, cases in which the first difference is smaller than the second difference and the diagnostic impact index is equal to or smaller than a second specified standard, as the presentation target cases;
The medical information processing device according to claim 2 . The diagnostic influence index has a larger value as the difference between the first diagnostic result and the diagnostic result of a doctor under the support of the clinical decision support system becomes smaller.
The medical information processing device according to claim 1 . a reception unit that receives a selection by a user of a setting of a first threshold value of the update influence index and a setting of a second threshold value of the diagnosis influence index;
a generation unit that generates a new case by duplicating a case of the plurality of cases included in the first data, the case having the update influence index equal to or greater than a first threshold and the diagnosis influence index equal to or greater than the second threshold, and adds the new case to the first data to generate the processed first data ,
the first data being training data to be used to update the clinical decision support system;
the second diagnosis result is a simulation result of a diagnosis for each of the plurality of diagnosed patients by the clinical decision support system after learning the first data;
The doctor's diagnosis results are diagnosis results for each of the plurality of diagnosed patients under the support of the clinical decision support system before the update.
The medical information processing device according to claim 1 . a learning control unit that controls a learning process in the learning of the clinical decision-making support system so that a higher priority is given to an example of a case in which a difference between the first diagnosis result and a diagnosis result of the doctor under the support of the clinical decision-making support system is smaller among the multiple examples included in the processed first data.
The medical information processing device according to claim 5 . the learning control unit controls the learning process so as to minimize a difference between the second diagnosis result of the clinical decision support system and the true value in learning of the clinical decision support system.
The medical information processing device according to claim 6 . The information about the disease included in the first data includes a plurality of types of test results for each of the plurality of diagnosed patients;
The learning control unit is
classifying the plurality of types of test results into a first test result having a large influence on the change between the first diagnostic result and the second diagnostic result, and a second test result having a smaller influence on the change between the first diagnostic result and the second diagnostic result than the first test result;
inputting the first test results into a first model;
inputting the second test results into a second model;
a weighting coefficient is adjusted so that the weight of the output result of the first model is increased, and then a weighted average of the output result of the first model and the output result of the second model is calculated, and the calculated weighted average is output as a prediction result.
The medical information processing device according to claim 6 or 7. The case extraction unit
determining that the influence is greater as the degree of similarity between the first diagnosis result and the diagnosis result of the doctor under the support of the clinical decision support system increases;
extracting, from among the plurality of cases included in the first data, cases that have a greater influence on the diagnosis result of the doctor than a third prescribed standard as the presentation target case;
The medical information processing device according to claim 1 . an influence variable extraction unit for extracting, from among a plurality of types of test result items included in the disease information, test items having a high degree of importance as influence variables having a high degree of importance for updating the clinical decision support system;
the influential variable extraction unit extracts influential variables having a high degree of importance based on the variable importance output from a prediction model trained using labels indicating classifications of important or unimportant for the items of the multiple types of test results as correct answers;
The output control unit causes the output unit to output the presentation target example in association with the influencing variable,
Among the test items, test items related to cases in which a difference between the first diagnostic result and a diagnostic result of the doctor under the support of the clinical decision support system is equal to or less than a second prescribed standard and a difference between the first diagnostic result and the second diagnostic result is equal to or more than a fourth prescribed standard are important test items;
Among the test items, test items relating to cases in which a difference between the first diagnostic result and a diagnostic result of the doctor under the support of the clinical decision support system is greater than a second specified criterion, or a difference between the first diagnostic result and the second diagnostic result is less than a fourth specified criterion, are non-important test items.
The medical information processing device according to claim 1 . The acquisition unit acquires second data including information on a disease for each of a plurality of patients that has not yet been diagnosed by a doctor and the clinical decision support system;
The output control unit causes the output unit to output information indicating a likelihood that a diagnosis result of the clinical decision support system for the plurality of patients, which is included in the second data, will change before and after learning the first data.
The medical information processing device according to claim 1 . The information about the disease included in the first data or the second data includes multiple types of test results for each of the multiple diagnosed patients or each of the multiple patients for whom a diagnosis has not been performed,
the output control unit causes the plurality of cases included in the first data and the cases related to the plurality of patients for whom the diagnosis has not been performed included in the second data to be mapped and displayed on a plane defined by values of two types of test results among the results of the plurality of types of tests.
The medical information processing device according to claim 11. The output control unit outputs the presentation target cases when the clinical decision support system is updated under a plurality of different conditions on the same screen for each of the plurality of different conditions.
The medical information processing device according to claim 1 . The diagnostic influence index is a report of a doctor who diagnosed a patient of the presented case regarding the degree of influence of the clinical decision support system on the diagnosis.
The medical information processing device according to claim 1 . an acquisition unit that acquires first data used for training a clinical decision-making support system, second data including information on a doctor and a disease for each of a plurality of patients for which a diagnosis has not been made by the clinical decision-making support system, a first diagnosis result for each of the plurality of diagnosed patients by the clinical decision-making support system before training with the first data, a second diagnosis result for each of the plurality of diagnosed patients by the clinical decision-making support system after training with the first data, and a doctor's diagnosis result for each of the plurality of diagnosed patients ;
a case extraction unit that calculates an update influence index and a diagnosis influence index for each of a plurality of cases included in the first data based on the first diagnosis result, the second diagnosis result, and the doctor's diagnosis result, and extracts, from the plurality of cases included in the first data, cases in which the update influence index is equal to or greater than a first specified standard and the diagnosis influence index is equal to or less than a second specified standard, as presentation target cases;
an output control unit that outputs information indicating a likelihood that a diagnosis result of the clinical decision support system for the plurality of patients included in the second data will change before and after learning the first data to an output unit, and also outputs the presented cases to the output unit in association with a diagnosis influence index indicating the degree of influence that the clinical decision support system has on a doctor's diagnosis in the presented cases and an update influence index;
Equipped with
The first data includes a plurality of cases in which information on a disease for each of the plurality of diagnosed patients is associated with a true value representing a treatment outcome for each of the plurality of diagnosed patients;
the update influence index is an index based on a difference between the first diagnosis result and the true value and a difference between the second diagnosis result and the true value,
The diagnostic influence index is an index based on a difference between the first diagnostic result and the doctor's diagnostic result.
Medical information processing equipment.

Images